Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: RE: Bivariate time series regression

From   "Jacobs, David" <>
To   "''" <>
Subject   st: RE: RE: Bivariate time series regression
Date   Tue, 20 Nov 2012 18:16:49 +0000

Oops I meant to write 'corrections for serial correlation are not well developed for "count models" '

D. Jacobs

-----Original Message-----
From: [] On Behalf Of Jacobs, David
Sent: Tuesday, November 20, 2012 1:15 PM
To: ''
Subject: st: RE: Bivariate time series regression

Others probably will be able to comment better than I can, but if your dependent variable numbers are fairly large, I'd use OLS and correct for serial correlation in one of the standard ways with say an AR1 term.  

The problem is that corrections for serial correlation don't seem to be well developed, although this may (probably still the case) have changed since the source I base this claim on was published.  

Of course, if (?) your counts are poisson distributed as they often are, than you need to correct for this with either a log link or by logging the dependent variable.  The latter option probably isn't as optimal as a log link.

Dave J.

-----Original Message-----
From: [] On Behalf Of Lott, Jason
Sent: Tuesday, November 20, 2012 11:20 AM
Subject: st: Bivariate time series regression

Hello everyone,

This is my first Statalist posting, so please forgive me for any errors or confusion.

I have two variables for which I am assessing a potential association:

(1) [Dependent variable]: Age-adjusted melanoma incidence rates for years 1935 to 2007. This is count data.

(2) [Independent variable]: Yearly sunspot intensity numbers (a measure of yearly solar flare activity) for years 1935 to 2007. This is continuous data.

The hypothesis being tested is that peak sunspot activity is associated with a future increase in yearly age-adjusted melanoma incidence rates. Sunspot activity is cyclical, with peak activity occurring approximately every 11 years. Yearly age-adjusted melanoma incidence rates are proposed to depend on prior, lagged peaks.

I have performed some initial trend analyses of each variable, both of which appear to be stationary.

My questions:

(1) Is choice of an ARIMAX model in Stata, with yearly (and lagged) sunspot activity numbers as a dependent covariate appropriate?

(2) Does Stata have Poisson-based regression packages that can account for autocorrelated dependent and independent time series variables?

As a neophyte health services researcher, I do not typically work with time series data, and I apologize for my lack of sophistication. I am more than happy to provide any additional information that may be needed, and I appreciate any help or guidance that might be provided.

My goal is to account for both time series (and autocorrelated) aspects of both variables in a regression model, while also having the flexibility to introduce time-lags of my independent variable as additional covariates. 


Jason Lott

Jason P. Lott, MD MSHP
Post-Doctoral Fellow
Robert Wood Johnson Clinical Scholars Program & Department of Dermatology Yale University School of Medicine

*   For searches and help try:

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index